Identify the proposition inside the verb
Associated with, correlated with, differed, and predicted can describe statistical relations without asserting that changing one variable would change another. Caused, led to, improved, reduced, prevented, and affected usually express causal change. Temporal order and adjustment alone do not turn an association into a causal effect.
Causal interpretation depends on design and assumptions: intervention or exposure definition, exchangeability, positivity, consistency, temporal order, measurement, interference, attrition, and the analysis used to estimate the target effect.
Rewrite observational overclaims
In this observational analysis, exposure X was associated with outcome Y after adjustment for [measured covariates].
Participants with [exposure] had [difference] in outcome Y; unmeasured confounding and reverse causation remain possible.
These patterns report the relation and its boundary. They do not imply that adjusting for measured variables removes every source of bias.
Use causal language only with a causal contract
State the design, target effect, identification assumptions, and relevant diagnostics. Describe violations and sensitivity analyses. Even a randomised study needs attention to adherence, missing outcomes, interference, and the estimand. Search the Phrase Explorer and inspect the causal guard on each result.